1

Orthogonal Time Frequency Space Modulation – Part

I: Fundamentals and Challenges Ahead
Zhiqiang Wei, Member, IEEE, Shuangyang Li, Member, IEEE, Weijie Yuan, Member, IEEE, Robert
Schober, Fellow, IEEE, and Giuseppe Caire, Fellow, IEEE
(Invited Paper)
Abstract—This letter is the first part of a three-part tutorial on ms. Thus, one channel coherence interval can only accommodate
orthogonal time frequency space (OTFS) modulation, which is a at most 3 OFDM symbols, i.e., a pilot symbol needs to be inserted
promising candidate waveform for future wireless networks. This once every three OFDM symbols, causing a large overhead.
letter introduces and compares two popular implementations of
OTFS modulation, namely the symplectic finite Fourier transform Recently, a new waveform, namely orthogonal time frequency
(SFFT)- and discrete Zak transform (DZT)-based architectures. space (OTFS) modulation, was proposed to cater to high-mobility
Based on these transceiver architectures, fundamental concepts of applications and has attracted significant attention since its intro-
OTFS modulation, including the delay-Doppler (DD) domain, DD duction [4], [6]. In particular, for OTFS modulation, the infor-
arXiv:2209.05011v1 [cs.IT] 12 Sep 2022

domain information multiplexing, and its potential benefits, are

discussed. Finally, the challenges ahead for OTFS modulation are mation symbols are placed in the delay-Doppler (DD) domain,
highlighted. Parts II and III of this tutorial on OTFS modulation wherein the time-varying channels are quasi-static and sparse
focus on transceiver designs and integrated sensing and communi- [7]. Besides, DD domain information multiplexing simplifies the
cation (ISAC), respectively. coupling between the information symbols and the channels,
Index Terms—OTFS, delay-Doppler domain, SFFT, DZT. which facilitates efficient and reliable transceiver design. More
importantly, owing to the two-dimensional (2D) spreading inher-
I. I NTRODUCTION ent to OTFS modulation, each DD domain symbol experiences
Next-generation wireless networks are required to support new the entire time-frequency (TF) domain channel response across
applications and challenging deployment scenarios, including one OTFS frame, which enables OTFS to exploit the full degrees
mobile communications on board aircraft (MCA), low-earth-orbit of freedom (DoFs) of high-mobility channels. Moreover, delay
(LEO) satellites [1], high speed trains, unmanned aerial vehicles and Doppler are typical sensing parameters and the DD domain
(UAVs) [2], and vehicle-to-vehicle (V2V) networks [3]. In this channel directly mirrors the geometry of the scatterers in high-
context, high-mobility communications have become an essential mobility environments [3], [8]. Therefore, OTFS modulation is
technology for future wireless networks [4]. Wireless commu- regarded as a promising waveform for integrated sensing and
nications in high-mobility propagation environments experience communication (ISAC) systems [9].
fast time-varying channels, which are typically caused by the The research on OTFS modulation, including the derivation of
Doppler effect [5]. If the Doppler effect was not considered its input-output characteristics [10]–[13], performance analysis
for system design, wireless communications over high-mobility [14]–[17], transceiver design [18]–[22], and its interplay with
channels would suffer from significant performance degradation. other promising techniques [23], [24], is quickly evolving and
The widely-adopted orthogonal frequency division multiplexing attracts an increasing number of researchers. However, it might
(OFDM) modulation may not be able to support efficient and be challenging for new researchers entering this field to grasp the
reliable communications in high-mobility environments due to the relevant mathematical tools and abstract concepts. The authors of
severe inter-carrier interference (ICI) induced by Doppler spread. [4] have provided an overview of the basic concepts, transceiver
Besides, the limited coherence time of the time-frequency domain architectures, and waveforms of OTFS modulation, but did not
channel will incur a large channel estimation overhead for OFDM provide any mathematical details. This motivates this three-part
systems. For example, for an OFDM system operating at a carrier tutorial, which aims to provide an accessible introduction for new
frequency of fc = 3.5 GHz, employing a subcarrier spacing of researchers interested in OTFS modulation. In particular, Part I
∆f = 15 kHz, and supporting a relative velocity of v = 300 introduces two popular implementations of OTFS modulation,
km/h, the maximum Doppler shift is νmax = 972.22 Hz while reviews some related fundamental concepts, and discusses future
the OFDM symbol duration including a 20% cyclic prefix (CP) is research challenges. Part II provides an in-depth discussion of
80 ms. The channel’s coherence time is around 4νmax 1
= 257.14 OTFS transceiver design, which is a key research direction in
the OTFS literature. Part III is dedicated to OTFS-based ISAC
Z. Wei is with the School of Mathematics and Statistics, Xi’an Jiaotong systems, which are regarded as an enabling technology for the
University, Xi’an 710049, China (e-mail: [email protected]). sixth-generation (6G) wireless networks, and highlights a range
S. Li was with the School of Electrical Engineering and Telecommunications,
University of New South Wales, Sydney, NSW 2052, Australia, when this of general research directions for OTFS modulation.
letter was submitted. He is now with the Department of Electrical, Electronic, We note that there are alternative waveforms [25], [26], which
and Computer Engineering, University of Western Australia, Perth, WA 6009,
Australia (e-mail: [email protected]).
are closely related to OTFS modulation and can also facilitate
W. Yuan is with the Department of Electrical and Electronic Engineering, high-mobility communications. However, in this tutorial, we focus
Southern University of Science and Technology, Shenzhen 518055, China (e-mail: on OTFS modulation due to the rapidly growing interest in this
[email protected]).
R. Schober is with the Institute for Digital Communications (IDC), the
modulation format.
Friedrich-Alexander University Erlangen-Nuremberg, Erlangen 91054, Germany Notations: CM ×N , RM ×N , and ZM ×N denote the sets of
(e-mail: [email protected]). M × N matrices with complex, real, and integer entries, re-
G. Caire is with the Electrical Engineering and Computer Science Department, ∗
the Technische Universität Berlin, Berlin 10587, Germany (e-mail: caire@tu- spectively; (·) denotes the conjugate operation; (·)N denotes the
berlin.de). modulus operation with respect to (w.r.t.) N ; δ (·) and δ [·] are
 2
YDD l, k yTD l  nM  r t  hDD  ,
DZT ADC

TABLE I
g rx l , k  N OTATIONS FOR M AIN S YSTEM PARAMETERS .
OTFS demodulation
Notation Physical meaning Notation Physical meaning
HDD l, k, l0 , k0
 
hDD (τ, ν) Continuous DD domain channel response Discrete DD domain channel response
w l, k, l0 , k0 , li , ki


Sampling function α [l, k, li , ki ] Phase rotation term
DZ x [l, k] Discrete Zak transform of x Agtx grx (τ, ν) Ambiguity function

OTFS modulation and νi ∈ [−νmax , νmax ] denote the delay and Doppler shifts
X DD l , k  of the i-th path, respectively, where τmax and νmax denote
X TF  m, n  Multicarrier s t 
ISFFT the maximum delay and Doppler shifts in the considered high-
modulation
mobility propagation environment, respectively. Note that DD
hDD  , 
YDD l , k  domain channel responses fluctuate much slower than time-delay
YTF  m, n  Multicarrier r t 
SFFT domain or TF domain channel responses, since only a drastic
demodulation
change of the propagation path lengths or the moving speeds
OTFS demodulation would cause DD domain channel response variations.
Fig. 1. SFFT-based OTFS transceiver architecture. After passing through the LTV channel, the received signal at
the OTFSZreceiver
Z (RX) is given by
the continuous- and discrete-time delta functions, respectively. ∞ ∞
For clarity, we summarize the notations for the main system r (t) = hDD (τ, ν)ej2πν(t−τ ) s (t − τ ) dτ dν + z (t)
parameters adopted in this paper in Table I. −∞ −∞
X P
= hi s (t − τi ) ej2πνi (t−τi ) + z (t) , (4)
II. T WO I MPLEMENTATIONS OF OTFS M ODULATION i=1
where z (t) denotes the additive white Gaussian noise (AWGN)
The two most commonly used implementations of OTFS mod- in the time domain. At the OTFS RX, after receive filtering, we
ulation are symplectic finite Fourier transform (SFFT)- [10], [11] first perform multicarrier demodulation of r (t) to obtain the TF
and discrete Zak transform (DZT)-based OTFS [12], [13]. In this domain received signal. This operation is also known as Wigner
section, we review the details of both transceiver architectures transform [10] andZ given by
∞
and discuss their similarities and differences. YTF [m, n] = ∗
r (t) grx (t − nT ) e−j2πm∆f (t−nT ) dt, (5)
A. SFFT-based OTFS −∞
where grx (t) is the RX filter. Then, the TF domain received signal
The SFFT-based OTFS transceiver architecture is shown in is transformed into the DD domain by the SFFT. The resulting
Fig. 1 [10], [11]. We assume a bandwidth of BOTFS and a time DD domain received signal is given by
N −1 M −1
duration of TOTFS , defined in the TF domain, are available for 1
YTF [m, n] e−j2π( N − M ) . (6)
X X kn lm

accommodating one OTFS frame. Bandwidth BOTFS is divided YDD [l, k]= √
N M n=0 m=0
into M subcarriers with subcarrier spacing ∆f = BOTFSM and time Note that (2) and (5) degenerate to conventional OFDM modula-
duration TOTFS is divided into N time slots with slot duration tion and demodulation, respectively, when N = 1 and rectangular
T = TOTFS
N . At the transmitter (TX) side, the information symbol TX and RX filters are adopted. Moreover, when N = 1 and no
for the (l, k)-th DD resource element (DDRE) is denoted as restrictions on the form of the TX and RX filters are imposed,
XDD [l, k], where l ∈ {0, . . . , M − 1} is the delay index and (2) and (5) degenerate to pulse-shaped OFDM modulation and
k ∈ {0, . . . , N −1} is the Doppler index. The DD domain symbols demodulation, respectively [28].
XDD [l, k] are transformed into TF domain signal XTF [m, n] via Combining (1)-(6), the DD domain input-output relationship
the inverse symplectic finite Fourier transform (ISFFT) [19], i.e., for OTFS modulation is given by
N −1 M −1 N −1 M −1
1 X X
XDD [l0 , k 0 ]HDD [l, k, l0 , k 0 ] + ZDD [l, k] , (7)
XDD [l, k] ej2π( N − M ) , (1)
X X nk ml
XTF [m, n] = √ YDD [l, k]=
N M k=0 l=0 k0 =0 l0 =0
where m ∈ {0, . . . , M − 1} is the subcarrier index and n ∈ where ZDD [l, k] denotes the effective DD domain noise. The
{0, . . . , N −1} is the slot index. The OTFS transmit signal s (t) is discrete DD domain X channel HDD [l, k, l0 , k 0 ] is given by
P
0 0
obtained by performing multicarrier modulation on the TF domain HDD [l, k, l , k ]= hi w (l, k, l0 , k 0 , li , ki ) e−j2πνi τi , (8)
i=1
signal XTF [m, n], also known as Heisenberg transform [10], i.e., where ki = νi N T ∈ R and li = τi M ∆f ∈ R. We distinguish
N
X −1 M
X −1 between integer and fractional Doppler and delay scenarios based
s (t) = XTF [m, n] gtx (t − nT ) ej2πm∆f (t−nT ) , (2) on whether ki and li are integer or fractional numbers, respec-
n=0 m=0 tively [10]. In (8), w (l, k, l0 , k 0 , li , ki ) is the sampling function
where gtx (t) is the TX pulse shaping filter. In practice, the for analog DD domain channel response, which captures joint
TF domain signal XTF [m, n] is first transformed to a time effects of delay and Doppler shifts, TX pulse shaping filter, and
domain sequence via the inverse finite Fourier transform (IFFT), RX filter, and is given by (9) at the top of next page, where the
is then pulse shaped in the discrete time domain, and is finally ambiguity function Agtx grx (τ, ν) is defined as
Z ∞
transformed to the analog TX signal s (t) by digital-to-analog ∗
conversion (DAC). Agtx grx (τ, ν) = gtx (t) grx (t − τ ) e−j2πν(t−τ ) dt. (10)
−∞
A deterministic channel model in the DD domain has been For the widely adopted rectangular TX and RX filters and
widely adopted for linear time-varying (LTV) channels [5], [27] assuming integer ki and li [10], (7) simplifies to the well-known
and the continuous DD domain channel response is modeled as 2D circular convolution input-output relationship as follows:
XP XP
hDD (τ, ν) = hi δ (τ − τi ) δ (ν − νi ) , (3) YDD [l, k] = hi XDD [(l−li )M , (k −ki )N ] α [l, k, li , ki ]
i=1 i=1
where P ∈ Z is the number of resolvable paths and hi ∈ C is + ZDD [l, k] , (11)
the channel coefficient of the i-th path. Variables τi ∈ [0, τmax ] where α [l, k, li , ki ] is a phase rotation term defined in (12) at the
top of next page.
 3

  
1 XN −1 XM −1 XM −1 1 1
w (l, k, l0 , k 0 , li , ki ) = A gtx grx −l i , (m − m 0
) ∆f − k i
NM n=0 m=0 m0 6=m M ∆f NT
  0
k0
XM −1 1 1 n(k−k −ki ) (ml−m l −m0 li )
0 0

+e−j2π N Agtx grx T − li , (m − m0 ) ∆f − ki e−j2π N ej2π M (9)

m0 =0 M ∆f NT

(l−li )M ki
ej2π M N l ∈ {li , . . . , M − 1}

α [l, k, li , ki ] = . (12)
 ej2π (l−li )M ki −kMi MN−(k−ki )N M l ∈ {0, . . . , l − 1} i

g tx l , k  OTFS modulation where DZ yTD [l, k] is the DZT of yTD [l + nM ] and DZ grx [l, k]
is the
 DZT of the periodically extended sampled RX filter
X DD l , k 
xTD l  nM  s t  grx [k]M MN
T , k ∈ Z.
IDZT DAC
A common choice for the TX and RX filters are rectangular
hDD  ,  1
YDD l , k 
filters. In this case, we have DZ gtx [l, k] = DZ grx [l, k] = √M ,
yTD l  nM  r t  N
DZT ADC ∀l, k. Moreover, assuming both integer delays and integer Doppler
shifts, the DD domain input-output relationship can be obtained
g rx l , k  OTFS demodulation as [13] XP
YDD [l, k] = hi XDD [(l−li )M , (k −ki )N ] α [k, l, ki , li ]
Fig. 2. DZT-based OTFS transceiver architecture. i=1
+ ZDD [l, k] , (19)
B. DZT-based OTFS
which is identical to (11). We refer to [13] for a more detailed
OTFS modulation
SFFT-based
X l , k  OTFSX modulation maps the DD domain informa- discussion of the input-output relationship for non-rectangular TX
DD
 m, n Multicarrier
TF s t 
tion symbols first ISFFTto the TF domain and then to the time domain
modulation and RX filters, fractional delays, and fractional Doppler shifts.
[10], [11]. In contrast, DZT-based OTFS modulation h  , 
directly
C. Their Similarities and Differences
Y l , k  the DD domain information symbols into the time
DD
transforms
DD
Y  m, n  r t 
Multicarrier
TF
The SFFT- and DZT-based OTFS implementations are both
domain, as shown SFFT in Fig. 2 [12], [13]. The DZT and inverse
demodulation
DZT (IDZT) define aOTFS mapping between a one-dimensional (1D) suitable for high-mobility communications. Besides, for rectan-
demodulation
sequence and a 2D sequence. Their definitions are provided in gular TX and RX filters, integer delays, and integer Doppler
the following. shifts, SFFT- and DZT-based OTFS lead to the same input-output
Definition 1 (Discrete Zak Transform): Let x be a periodic relationship, see (11) and (19). Unveiling their connections for the
sequence with period M N . The DZT of x is defined as case of non-rectangular filters, fractional delays, and fractional
∆ 1 XN −1 n
Doppler shifts is still an open problem in the literature.
DZ x [l, k] = √ x [l + nM ] e−j2π N k , (13) SFFT-based OTFS is compatible with conventional OFDM by
N n=0
concatenating the ISFFT/SFFT module and an OFDM modula-
for l ∈ {0, . . . , M − 1} and k ∈ {0, . . . , N − 1}.
tor/demodulator, while DZT-based OTFS enjoys a lower computa-
Definition 2 (Inverse Discrete Zak Transform): Given the M N - tional complexity as only N -point inverse finite Fourier transform
periodic sequence x, whose DZT is given by DZ x [l, k], the IDZT (IFFT) and finite Fourier transform (FFT) are needed with a
is defined as symbol-wise interleaver/de-interleaver at the OTFS transceivers,
∆ 1 XN −1 n
x [l + nM ] = √ DZ x [l, k] ej2π N k , (14) respectively, see (15) and (18). Moreover, the SFFT-based OTFS
N k=0
architecture facilitates TF domain signal processing, such as
for l ∈ {0, . . . , M − 1} and n ∈ {0, . . . , N − 1}. window function design [19] and design of multiple access
As shown in [13], a time-delay (TD) domain sequence can be schemes [29]. On the other hand, for DZT-based OTFS, the TX
obtained by applying the IDZT to the DD domain symbols at the and RX filters can be implemented digitally in the DD domain
OTFS TX, i.e., based on their DZTs. This not only reduces the implementation
√ XK−1 n
xTD [l + nM ] = M XDD [l, k] DZ gtx [l, k] ej2π N k , complexity, but also facilitates a systematic pulse shape design in
k=0
(15) the DD domain. A detailed discussion on pulse shape design can
where DZ gtx [l, k] is the DZT  of the periodically extended be found in Part II of this tutorial.
sampled TX pulse shape gtx [k]M MN
T , k ∈ Z. The OTFS
III. D ISCUSSIONS ON OTFS M ODULATION
transmit signal s(t) can be obtained by DAC of xTD [l + nM ].
Interpolating with a rectangular kernel, the OTFS transmit signal Based on the transceiver architectures introduced in Section II,
is given by we shed light on some fundamental concepts of OTFS modulation
  in this section.
l−0.5 l+0.5
s(t) = xTD [l+nM ] , t ∈ T +nT, T +nT . (16) A. What is the Delay-Doppler Domain
M M
At the OTFS RX, the channel output r(t) in (4) is subjected In the context of signal processing, we define and describe a
to an analog-to-digital conversion (ADC) and the resulting TD signal in a certain domain. Describing a signal in the time domain
domain received signal is given by characterizes the fluctuation of its amplitude and phase w.r.t. the
time variable, while representing a signal in the frequency domain
yTD [l + nM ] = r(t)|t= l
T +nT . (17)
M characterizes the corresponding fluctuation rate. Furthermore, the
Then, the DD domain received signal is obtained by performing TF domain is expanded by the time duration and frequency
DZT on the TD domain received signal, i.e., bandwidth occupied by a signal. The characterization of a signal
√
YDD [l, k] = M N DZ yTD [l, k] DZ ∗grx [l, k] , (18) in the TF domain reveals the temporal evolution of its spectral
 4

content (within a certain time duration). On the other hand, the IV. C HALLENGES A HEAD
delay and Doppler parameters are usually used to characterize a
So far, the research on OTFS modulation has focused on
time-varying system rather than a signal. For example, h (τ, ν)
the characterization of the input-output relationship, transceiver
in (3) characterizes the response of a time-varying system that
design, and potential synergies with other communication and
introduces delays τi , ∀i, and Doppler shifts νi , ∀i, to the input
signal processing techniques, as elaborated in Parts I, II, and
signal. OTFS modulation is based on signal representation theory
III, respectively. However, some aspects of OTFS modulation
[7], cf. (1), (2), and (15), which allows to characterize a signal
are not well understood or have not been addressed by the
directly in the DD domain. In particular, a DD domain impulse
research community, yet. In this section, we discuss important
δ [l − l0 , k − k0 ], l, l0 ∈ {0, . . . , M − 1}, k, k0 ∈ {0, . . . , N − 1}
open problems, which might be the main challenges ahead for
corresponds to a time-varying pulsetone in the time domain [7],
OTFS research.
since it can be viewed as the output of a virtual time-varying
system that introduces delay l0 M1∆f and Doppler shift k0 N1T A. Latency Concern
w.r.t. an input DD domain impulse δ [l, k]. At the OTFS RX,
when transforming a signal from the time or TF domain to Compared to OFDM, OTFS modulation spreads information
the DD domain, cf. (5), (6), and (18), we are inverting the over a longer block to combat channel fluctuations, which may
channel operations the TX signal has experienced. According to increase the overall latency. In particular, the duration of each
the uncertainty principle of the Fourier transform, the longer the OTFS signal is N times of that of an OFDM symbol, with the
TX signal duration, the higher the resolution for distinguishing same number of subcarriers and the same subcarrier spacing [10].
different Doppler shifts in the channel, while the larger the Fortunately, multiplexing information in the DD domain turns
TX signal frequency bandwidth, the higher the resolution for a time-varying channel into a quasi-static channel in the DD
distinguishing different delay shifts in the channel [12]. domain, which might significantly reduce the amount of signal-
ing overhead required for channel estimation. A comprehensive
B. How to Embed Information into the DD Domain comparative study of the latency of OTFS and OFDM is still
As shown in (15), the OTFS TX modulates information sym- missing. Besides, the latency of OTFS can be made identical to
bols XDD [l, k] onto the DD domain pulse samples DZ gtx [l, k], that of OFDM by increasing the subcarrier spacing or reducing
l ∈ {0, . . . , M −1}, k ∈ {0, . . . , N −1}, which can be interpreted N at the expense of a reduced Doppler resolution. A detailed
as the generation of a set of time-varying carrier signals in the performance comparison of OTFS and OFDM for a given latency
time domain to carry the corresponding information symbol. A is an interesting research direction.
set of 2D impulses with zero bandwidth in the DD domain, i.e.,
δ [l − l0 , k − k0 ], l, l0 ∈ {0, . . . , M − 1}, k, k0 ∈ {0, . . . , N − 1}, B. Fractional Doppler
would be ideal for carrying information since they are perfectly To limit the latency of OTFS modulation, the signal duration
localized [12]. Unfortunately, this is not possible in practice since N T cannot be chosen exceedingly large, which gives rise to the
such impulses would require infinite time duration and infinite well-known fractional Doppler issue [10]. Existing works in the
frequency bandwidth. As shown in [12], 2D sinc pulses (their literature have addressed the DD domain channel estimation prob-
samples are given by DZ gtx [l, k] in (15), l ∈ {0, . . . , M − 1}, lem in the case of fractional Doppler by various means, e.g., using
k ∈ {0, . . . , N −1}) with finite time duration and frequency band- windowing [19], oversampling [30], and off-grid estimation [20].
width, which are approximately orthogonal in the DD domain, However, detection performance may significantly degrade in the
constitute an efficient set of DD domain information-carrying case of fractional Doppler. In fact, the commonly used message
impulses. passing-based detection algorithms [10] are prone to diverge, due
to the loops in the factor graph caused by fractional Doppler shift.
C. Benefits of DD Domain Modulation
How to improve detection performance and robustness in the case
Due to the Heisenberg uncertainty principle, it is impossible of fractional Doppler is a critical issue. We note that the authors in
to find a 2D signal perfectly localized in the TF domain, while [21], [31] have proposed a cross domain message passing detector
in the DD domain, we can obtain a set of 2D localized sig- and a variational Bayes detector, respectively, which can improve
nals (their samples are given by DZ gtx [l, k] in (15)) that are the convergence performance in the case of fractional Doppler.
approximately orthogonal [12]. Moreover, as mentioned before,
the OTFS modulator employs time-varying carrier signals for C. Multiple Access
carrying information, which caters to the time-varying features of
Serving multiple users in high-mobility environments is a
high-mobility channels. In particular, the high-mobility channel,
challenging task. For OTFS modulation, it is not clear yet in
corresponding to DD domain channel response, hDD (τ, ν) in (3)
which domain the user should be multiplexed. Conventional
can be modeled as a linear time-varying system concatenated with
TF domain multiple access schemes may suffer from severe
the virtual time-varying system formed by the OTFS modulator.
multi-user interference (MUI) caused by doubly selective fading
As a result, the high-mobility channel additively changes the time-
channels. On the other hand, DD domain multiple access requires
varying features of the carrier signals, i.e., the received carrier
careful allocation of the DDREs and the guard space between
signal experiences a shorter or a longer delay (l±li ) and a lower or
users to minimize MUI. Moreover, the 2D circular convolution
a higher Doppler shift (k±ki ), compared to the transmitted carrier
input-output relationship of OTFS leads to special MUI patterns
signal in the (l, k)-th DDRE. As a result, the additive channel
that might be exploited for user scheduling and resource allocation
operations on the carrier signals, i.e., the delay and Doppler shifts,
design. Furthermore, compared to orthogonal multiple access,
can be simply described by a shift operation and a phase rotation
power domain and/or code domain non-orthogonal multiple ac-
in the DD domain, i.e., (11) and (19), which is beneficial for
cess schemes [32] can accommodate more users while imposing
efficient and reliable channel equalization.
a critical challenge for interference management.
 5

D. MU MIMO-OTFS [6] R. Hadani, S. Rakib, M. Tsatsanis, A. Monk, A. J. Goldsmith, A. F. Molisch,

and R. Calderbank, “Orthogonal time frequency space modulation,” in Proc.
Multiple-input multiple-output (MIMO) systems can provide IEEE Wireless Commun. and Networking Conf., 2017, pp. 1–6.
additional spatial DoFs for OTFS systems, which may be ex- [7] R. Hadani and A. Monk, “OTFS: A new generation of modulation addressing
ploited to improve communication reliability and the achievable the challenges of 5G,” arXiv preprint arXiv:1802.02623, 2018.
[8] L. Gaudio, M. Kobayashi, G. Caire, and G. Colavolpe, “On the effectiveness
rate. Moreover, these additional DoFs can be used to accommo- of OTFS for joint radar parameter estimation and communication,” IEEE
date multiple users in the spatial domain, which would allow Trans. Wireless Commun., vol. 19, no. 9, pp. 5951–5965, Sep. 2020.
to avoid the guard resource elements needed in existing DD and [9] S. Li, W. Yuan, C. Liu, Z. Wei, J. Yuan, B. Bai, and D. W. K. Ng, “A novel
ISAC transmission framework based on spatially-spread orthogonal time
TF domain multiple access schemes [18]. However, incorporating frequency space modulation,” to appear in IEEE J. Select. Areas Commun.,
OTFS modulation in MU MIMO systems is a challenging task 2022.
since the signals have to be optimized simultaneously in the delay, [10] P. Raviteja, K. T. Phan, Y. Hong, and E. Viterbo, “Interference cancellation
Doppler, and spatial domains. The authors of [33] investigated the and iterative detection for orthogonal time frequency space modulation,”
IEEE Trans. Wireless Commun., vol. 17, no. 10, pp. 6501–6515, Oct. 2018.
precoding design for massive fully-digital MIMO-OTFS systems, [11] A. Farhang, A. RezazadehReyhani, L. E. Doyle, and B. Farhang-Boroujeny,
which may incur exceedingly high hardware complexity and “Low complexity modem structure for OFDM-based orthogonal time fre-
power consumption. In fact, a general and concise MU MIMO- quency space modulation,” IEEE Commun. Lett., vol. 7, no. 3, pp. 344–347,
Jun. 2018.
OTFS model, based on which channel estimation, precoding, and [12] S. K. Mohammed, “Derivation of OTFS modulation from first principles,”
multiple access schemes can be designed, has not been reported IEEE Trans. Veh. Technol., vol. 70, no. 8, pp. 7619–7636, Aug. 2021.
so far. [13] F. Lampel, A. Alvarado, and F. M. Willems, “Orthogonal time frequency
space modulation: A discrete Zak transform approach,” arXiv preprint
E. Cross-layer Design arXiv:2106.12828, 2021.
[14] G. D. Surabhi, R. M. Augustine, and A. Chockalingam, “On the diversity
High-mobility communication networks, such as UAV and of uncoded OTFS modulation in doubly-dispersive channels,” IEEE Trans.
V2V networks, may suffer not only from time-varying channels Wireless Commun., vol. 18, no. 6, pp. 3049–3063, Jun. 2019.
but also from a highly dynamic network topology. However, [15] S. Li, J. Yuan, W. Yuan, Z. Wei, B. Bai, and D. W. K. Ng, “Performance
analysis of coded OTFS systems over high-mobility channels,” IEEE Trans.
existing works on OTFS modulation mainly focus on the physical Wireless Commun., vol. 20, no. 9, pp. 6033–6048, Sep. 2021.
layer design to improve communication performance. Cross-layer [16] A. RezazadehReyhani, A. Farhang, M. Ji, R. R. Chen, and B. Farhang-
design spanning both the physical and network layers for OTFS Boroujeny, “Analysis of discrete-time MIMO OFDM-Based orthogonal time
frequency space modulation,” in Proc. IEEE Intern. Commun. Conf., May
systems is a critical but challenging task. In fact, cross-layer 2018, pp. 1–6.
design facilitates coordination, interaction, and joint optimization [17] R. Chong, S. Li, J. Yuan, and D. W. K. Ng, “Achievable rate upper-bounds of
of communication protocols across different layers, which is ben- uplink multiuser OTFS transmissions,” to appear in IEEE Wireless Commun.
Lett., 2022.
eficial for mobility management and energy-efficient and reliable
[18] P. Raviteja, K. T. Phan, and Y. Hong, “Embedded pilot-aided channel
communications. In high-mobility environments, it is critical to estimation for OTFS in delay-Doppler channels,” IEEE Trans. Veh. Technol.,
sense the fast time-varying network state information (NSI), such vol. 68, no. 5, pp. 4906–4917, May 2019.
as the location and velocity of the network elements exploiting the [19] Z. Wei, W. Yuan, S. Li, J. Yuan, and D. W. K. Ng, “Transmitter and receiver
window designs for orthogonal time-frequency space modulation,” IEEE
OTFS modulation waveform, and to design the OTFS transceivers Trans. Commun., vol. 69, no. 4, pp. 2207–2223, Apr. 2021.
accordingly. Furthermore, the network topology may be actively [20] ——, “Off-grid channel estimation with sparse Bayesian learning for OTFS
adjusted depending on the state of the high-mobility environment systems,” to appear in IEEE Trans. Wireless Commun., 2022.
[21] S. Li, W. Yuan, Z. Wei, and J. Yuan, “Cross domain iterative detection
to improve the communication performance. for orthogonal time frequency space modulation,” IEEE Trans. Wireless
Commun., vol. 21, no. 4, pp. 2227–2242, Apr. 2022.
V. C ONCLUSIONS [22] Z. Yuan, F. Liu, W. Yuan, Q. Guo, Z. Wang, and J. Yuan, “Iterative detection
for orthogonal time frequency space modulation with unitary approximate
As the first part of this three-part tutorial, this letter presented message passing,” IEEE Trans. Wireless Commun., vol. 21, no. 2, pp. 714–
and compared two common implementations of OTFS, namely 725, Jul. 2022.
SFFT- and DZT-based OTFS. Based on the introduced OTFS [23] Z. Ding, R. Schober, P. Fan, and H. Vincent Poor, “OTFS-NOMA: An
efficient approach for exploiting heterogenous user mobility profiles,” IEEE
transceiver architectures, some fundamental concepts of OTFS Trans. Commun., vol. 67, no. 11, pp. 7950–7965, Nov. 2019.
modulation were discussed. Subsequently, several important re- [24] L. Gaudio, M. Kobayashi, G. Caire, and G. Colavolpe, “Hybrid digital-
search challenges regarding OTFS modulation were highlighted. analog beamforming and MIMO radar with OTFS modulation,” arXiv
Part II and Part III will discuss OTFS transceiver design and the preprint arXiv:2009.08785, 2020.
[25] T. Thaj, E. Viterbo, and Y. Hong, “Orthogonal time sequency multiplexing
benefits of OTFS for ISAC and other applications, respectively. modulation: Analysis and low-complexity receiver design,” IEEE Trans.
Wireless Commun., vol. 20, no. 12, pp. 7842–7855, Dec. 2021.
[26] T. Zemen, M. Hofer, D. Löschenbrand, and C. Pacher, “Iterative detection
R EFERENCES for orthogonal precoding in doubly selective channels,” in Proc. IEEE
[1] N. Cheng, J. He, Z. Yin, C. Zhou, H. Wu, F. Lyu, H. Zhou, and X. Shen, “6G Personal, Indoor and Mobile Radio Commun. Sympos., Sep. 2018, pp. 1–7.
service-oriented space-air-ground integrated network: A survey,” to appear [27] Y. Hong, T. Thaj, and E. Viterbo, Delay-Doppler Communications: Princi-
in Chinese Journal of Aeronautics, Dec. 2021. ples and Applications. Academic Press, 2022.
[2] Y. Cai, Z. Wei, R. Li, D. W. K. Ng, and J. Yuan, “Joint trajectory and [28] T. Strohmer and S. Beaver, “Optimal OFDM design for time-frequency
resource allocation design for energy-efficient secure UAV communication dispersive channels,” IEEE Trans. Commun., vol. 51, no. 7, pp. 1111–1122,
systems,” IEEE Trans. Commun., vol. 68, no. 7, pp. 4536–4553, Mar. 2020. Jul. 2003.
[3] W. Yuan, Z. Wei, S. Li, J. Yuan, and D. W. K. Ng, “Integrated sensing and [29] V. Khammammetti and S. K. Mohammed, “OTFS-based multiple-access in
communication-assisted orthogonal time frequency space transmission for high Doppler and delay spread wireless channels,” IEEE Wireless Commun.
vehicular networks,” IEEE J. Select. Topics Signal Process., vol. 15, no. 6, Lett., vol. 8, no. 2, pp. 528–531, Apr. 2019.
pp. 1515–1528, Nov. 2021. [30] Y. Ge, Q. Deng, P. C. Ching, and Z. Ding, “Receiver design for OTFS with
[4] Z. Wei, W. Yuan, S. Li, J. Yuan, G. Bharatula, R. Hadani, and L. Hanzo, a fractionally spaced sampling approach,” IEEE Trans. Wireless Commun.,
“Orthogonal time-frequency space modulation: A promising next-generation vol. 20, no. 7, pp. 4072–4086, Jul. 2021.
waveform,” IEEE Wireless Commun., vol. 28, no. 4, pp. 136–144, Aug. 2021. [31] W. Yuan, Z. Wei, J. Yuan, and D. W. K. Ng, “A simple variational Bayes
[5] F. Hlawatsch and G. Matz, Wireless Communications Over Rapidly Time- detector for orthogonal time frequency space (OTFS) modulation,” IEEE
varying Channels. Academic Press, 2011. Trans. Veh. Technol., vol. 69, no. 7, pp. 7976–7980, Jul. 2020.
 6

[32] Z. Wei, L. Zhao, J. Guo, D. W. K. Ng, and J. Yuan, “Multi-beam NOMA

for hybrid mmwave systems,” IEEE Trans. Commun., vol. 67, no. 2, pp.
1705–1719, Feb. 2019.
[33] B. C. Pandey, S. K. Mohammed, P. Raviteja, Y. Hong, and E. Viterbo, “Low
complexity precoding and detection in multi-user massive MIMO OTFS
downlink,” IEEE Trans. Veh. Technol., vol. 70, no. 5, pp. 4389–4405, May
2021.